Networking Report (NWR) Subscribe

CES Sees New MU-MIMO Chipsets

Vendors Rush to Support 802.11ac Wave 2 Access Points

February 2, 2015

By Loring Wirbel

Leading Wi-Fi chipset vendors touted their next-generation MU-MIMO platforms at the recent Consumer Electronics Show. These companies are anxious to support an anticipated rush of Wave 2 access points (APs) for the 802.11ac standard; among them were Broadcom, Marvell, Qualcomm Atheros, and Quantenna.

In 2013, Broadcom touted its XStream technology for 802.11ac access points while preparing its multiuser MIMO solution. Now, by sampling the BCM4366 for 4x4 MU-MIMO, the company is taking a dual-track approach. Even as XStream continues to find favor in lower-cost dual-band modems, Broadcom is supporting full 4x4 MU-MIMO for Wave 2 devices. The BCM4366 can operate on its own in gateways and set-top boxes. It integrates an ARM Cortex-A7 CPU and can also serve with a companion BCM47094 SoC (an enhanced version of the proven BCM4709) for retail routers and gateways as well as more-demanding enterprise routers. The BCM47094 combines a dual-core 1.4GHz ARM Cortex-A9 with an Ethernet switch and USB3.0 port.

Cisco says 802.11ac’s multiuser feature for Wave 2 is “switch-like” from the client’s perspective, because it enables up to four clients to achieve full bandwidth simultaneously. The AP transmits several data streams to multiple clients at the same time, using advanced beam forming to send only one spatial stream to a given user, if desired. It could send aggregated streams to a particular client for a 1.7Gbps throughput using 80MHz channels. A single-antenna client can only receive data at 433Mbps, but a four-antenna AP could use MU-MIMO beam steering to send different data streams to each client.

In a recent whitepaper discussing its Aironet 3600 AP, Cisco introduced a nomenclature that includes spatial streaming: for instance, 4x4:3 represents four transceivers and antennas supporting only three spatial streams. In agreement with all four MU-MIMO chipset vendors, the company says throughput can vary significantly for a given transceiver/antenna/stream architecture. Given configurations rarely if ever achieve the theoretical maximum.

Quantenna was first to sample a true 4x4 MIMO chipset, basing its design on the older QHS710 for 802.11n MIMO systems. The QSR1000 reached production in 3Q13.

In 1H14, when Qualcomm Atheros introduced its first MU-MIMO designs, Broadcom countered with a two-radio solution called XStream, which uses two 5GHz radios and a third for 2.4GHz. Broadcom’s SmartConnect software enables user devices to roam seamlessly among the radios, maximizing Wi-Fi network bandwidth (see NWR 4/28/14, “Vendors Tussle Over Multiuser Wi-Fi”). Now that the Wi-Fi Alliance is readying its first certification efforts for 802.11ac Wave 2, Broadcom has jumped into the fray with true MU-MIMO support. But the company anticipates a healthy market for XStream, which it can implement in a six-stream version using the BCM3602 and an eight-stream version using the newer BCM4366.

Marvell and Quantenna were among the CES vendors hoping to accelerate OEM interest in MU-MIMO chipsets for Wave 2 designs. Marvell says its Avastar 86W8964 is the first true 4x4 MIMO chipset to support 160MHz channels and 2.6Gbps speeds along with 802.11mc precision-location features. Quantenna showed prototypes of its promised QSR10G 10Gbps design implementing 8x8 MU-MIMO, but it didn’t announce sampling. The company has gained important router customers such as Linksys for its 4x4 design, however, and it demonstrated a new router reference design in collaboration with Lantiq.

Searching for a Wave

Although 1080p and 4K video streams over wireless 802.11ac links remain an immediate need for some home-networking vendors, Wave 1 suffices for the bulk of consumer applications. Many enterprise applications could take advantage of immediate Wave 2 upgrades, however. CES activity focused less on end-user demands than on preparing Wave 2 products for testing before the certifications begin. The Wi-Fi Alliance (WFA) promised a 2H14 certification program as the IEEE standard moved to completion, but no such program emerged in 2014. The WFA initially promoted Wave 1 as a simpler means of bringing products to market; this initial wave supported three data streams in up to 3x3 configurations, but not MU-MIMO.

Although OEMs can sell products without certifications, the WFA seal of approval is a virtual necessity for corporate sales. Wave 1 tests began in 2013 and covered routers and access points that implemented 802.11ac using 20MHz, 40MHz, and 80MHz channels in the 5GHz band. Those APs, however, lacked all the beam-forming features of the newer multiuser-MIMO chipsets (see NWR 8/12/13, “Wi-Fi Alliance Blesses 802.11ac”). The 802.11ac standard allows up to eight independent spatial streams; in most cases, though, only Wave 2 modems are expected to use eight streams. In fact, they are the first modems to add a standard fourth spatial stream.

Wave 2, covering 4x4 MIMO, MU-MIMO, and 160MHz channels, represents 802.11ac’s true differentiation potential in both throughput and configuration flexibility, offering a maximum data rate of 6.96Gbps. OEMs at CES expect WFA certification procedures to begin before 2H15 and were waiting to see which IC suppliers could promise early production. The WFA may begin informal testing of some products in early February, and it has planned the first of multiple plugfests for April, with additional plugfests tentatively slated for June, August, and October. Because formal certifications usually follow the full series of plugfests, none is likely before 1H16.

Many 802.11ac features are optional, including MU-MIMO. Wave 2 allows but does not mandate 160MHz contiguous or 80MHz+80MHz noncontiguous channels. (The standard requires that an 80Mz channel comprise two adjacent nonoverlapping 40MHz channels.) Higher-order-modulation rates such as QAM-256 are also optional. Vendors have touted support for features that enterprise applications require, offering proof of their commitment to Wave 2 while suggesting that chipsets without these features are not full-featured 802.11ac products.

Employing 160MHz contiguous channels can be problematic because the FCC has allowed weather radar to continue using some 5GHz channels. The 802.11ac chips must therefore perform dynamic frequency selection (DFS) to avoid existing radar transmissions. An end-user modem can achieve DFS certification rather easily, but demonstrating automatic backoff from a channel that carries an interfering radar transmission can be tougher. Ideally, vendors should offer DFS stability, which offers accurate signal detection with few false positives, but switching to another channel still allows carrier-grade streaming video to be transmitted without errors.

For consumers and small offices, 160MHz channels offer a low-cost path to greater throughput; a 2x2 MIMO design using two spatial streams and a 160MHz channel delivers the same maximum data rate as a 4x4 MIMO design using four spatial streams and an 80MHz channel (assuming both employ the same modulation). The former requires fewer RF components and less processing, reducing both bill-of-material and chipset costs.

Enterprise applications, however, need more nonoverlapping channels to support adjacent APs, so they favor 80MHz. In this environment, 4x4 MIMO with four spatial streams offers both high throughput and an adequate number of channels. We note, however, that some vendors are marketing designs that support only three independent spatial streams as being 4x4 MIMO. Such 4x4:3 designs use four antennas for beam forming but are limited to only three-quarters of the maximum throughput relative to a 4x4:4 implementation.

As Cisco points out in the aforementioned whitepaper, however, a system can be designed explicitly for three spatial streams, using a fourth transceiver for better MIMO equalization gain on the uplink and more-efficient beam forming on the downlink. Depending on the system-level design, a 4x4:3 AP could offer equivalent or even better typical performance compared with a 4x4:4 AP.

Challenges to an Early Leader

Qualcomm Atheros (QCA) was second only to Quantenna in offering diverse chipsets when it launched its product suite last April. The 9980 and 9990 support four antennas in an access point; the 9982 and 9992 are cost-reduced versions supporting three antennas. The 998x devices target residential access points, whereas the 999x devices are optimized for enterprise applications. QCA brands the 802.11ac family “Vive” and its proprietary extensions to standard MU-MIMO “MU/EFX.”

Access-point vendors demonstrating Qualcomm-based products at CES included Amped Wireless, Buffalo BBS, D-Link, NEC, TP-Link, and TrendNet. Client customers included Acer and Xiaomi. The company hinted at a 2H15 tri-band product that would combine 5GHz 802.11ac with MU-MIMO, 2.4GHz radios for backward compatibility, and the 60GHz (millimeter-wave) 802.11ad technology the company picked up by acquiring Wilocity. This last feature has yet to generate widespread support among potential customers owing to the common misconception that 60GHz components carry unit prices several times those of 2.4GHz and 5GHz radios. Emerging applications for in-room serial links and wireless backhaul, however, could generate the interest QCA anticipates for tri-band designs that include 802.11ad.

Broadcom must aggressively promote unique features if it’s to overcome Quantenna’s and QCA’s early lead. It points to the advantages of retaining two independent radios, even while moving from XStream to MU-MIMO. For example, radio redundancy enables a 5GHz “zero-wait DFS” in which one radio chain can handle DFS detection while the other radio chains operate normally. Broadcom also is heavily promoting what it calls NitroQAM, a proprietary modulation scheme that extends to QAM-1,024 and can increase data rates at short range. The company says using NitroQAM on both links will support throughput of up to 2.133Gbps.

Both Broadcom and Marvell promote use of ARM-based processors alongside the 802.11ac SoC to improve the performance of higher-end enterprise and retail routers. The BCM4366 combines a radio core with MU-MIMO beam-forming logic and an ARM Cortex-A7 core, as Figure 1 shows, though it integrates only a PCIe interface, not an Ethernet port. Broadcom offers the BCM47094 with dual 1.4GHz ARM cores, while Marvell offers an AP3200 reference platform combining the Avastar 86W8964 with its dual-core 88F6282 Armada processor and support software. Marvell says Avastar is the first architecture to offer a true 160MHz channel, though its 4x4 design only supports three spatial streams for a peak data rate of 2.6Gbps.

Figure 1. Broadcom’s BCM4366 integrates an ARM Cortex-A7 CPU. Broadcom touts the BCM4366 for either 4x4 MU-MIMO or the 5GHz XStream designs it introduced last spring. XStream, however, may only find a place in legacy designs. (Source: Broadcom)

Broadcom thinks retail customers will be the main users of the ARM-based BCM47094 processor. Marvell estimates that all retail and more than 90% of enterprise customers will rely on Armada-assisted designs. At the same time, the company aims to sell 802.11ac chipsets at a range of prices by offering 1x1 and 2x2 products before the year ends.

Qualcomm Atheros is satisfied with the processing power of its 998/999x families, but some gateway designs may augment them by adding the IPQ Internet processors with dual Krait cores, which Qualcomm introduced in late 2013 (see NWR 12/9/13, “Qualcomm Unveils Comms Processors”). The company says that filling out the product line with lower-end antenna configurations for sub-$100 routers will likely be more important than adding processing power.

The 10Gbps End Game

Although Quantenna’s first-generation QSR1000 lacked the160MHz channels that Marvell offers, the company has attracted OEM attention through a promised 8x8 design with eight spatial data streams delivering 10Gbps. Quantenna brought no sample silicon to CES, but it still plans to have designs in production by 2016. In the meantime, the QSR1000 device for 4x4 MU-MIMO builds on the company’s QHS710 4x4 design originally developed for 802.11n. The 10Gbps chipset will support 160MHz channels, unlike the current 4x4 MU-MIMO QSR1000. Nevertheless, sporting 80MHz channels, it has appeared in such APs as the Asus RT-AC87U and Netgear Nighthawk X4.

While Quantenna risks souring customers on its 4x4 designs if it promotes future 8x8 configurations too heavily, competitors are wondering whether expanding beyond four spatial streams makes sense. Because 4x4 configurations have hit the power and real-estate sweet spot, examining other frequency bands could make more sense.

Given its acquisition of Wilocity, a pioneer in client-side millimeter-wave radio (see NWR 5/19/14, “Millimeter Wave Seeks Best Home”), Qualcomm Atheros’s demonstration of a tri-band access point that combines 802.11ac with 802.11ad at 60GHz was unsurprising. The company says a chicken-and-egg problem remains between the AP and client for 60GHz ubiquity, but in-room applications are emerging for 60GHz tri-band smartphones, which may help drive the multiband AP.

OEMs may have to weigh the benefits of increasing the number of spatial streams in traditional MU-MIMO 802.11ac APs versus moving to the newer 60GHz band, which can add 7Gbps of throughput. Many are surprised to find that a millimeter-wave radio IC does not cost five times more than a 5GHz device, but rather incurs only a small price difference similar to that of moving from 2.4GHz to 5GHz—a factor that may help drive tri-band access points.

Debating the Size of a New Wave

Some wireless access OEMs like Aruba and Ruckus have taken the slightly jaded marketing position that instead of waiting for Wave 2 APs, customers should buy Wave 1 products, since the real 802.11ac revolution came with the move to 5GHz frequencies and antenna diversity.

We see each step in the evolution of Wi-Fi networks as guaranteeing the planned obsolescence of the previous one, making both moves in 802.11ac critical. Just as 802.11n quickly soured the market for 802.11a/g, the arrival of Wave 1 products has eliminated all but entry-level 802.11n products. There is no reason to think Wave 1 products will avoid a similar obsolescence when Wave 2 arrives late this year.

Although Wave 2 has more options than mandated features, it represents the first big shift to MU-MIMO, to three and four spatial streams of data, and to 160MHz channels. Where chip vendors offer one feature but not another, as Table 1 shows, the differences look like application optimization rather than product limitations, since these features represent cost/performance tradeoffs.

Table 1.  Comparison of 802.11ac Wave 2 chipsets. Since many Wave 2 features are optional, the four major vendors implement different subsets, leading to a range of client throughput rates and AP performance. (Source: The Linley Group)

Each of the four major vendors of 802.11ac chips will insist that its choice of channel width and spatial-stream count is optimal, but they all agree that the real market confusion comes from a handful of Asian chip manufacturers trying to enter with devices that probably lack MU-MIMO support. We believe the four vendors currently offering products will remain the main players for enterprise and high-performance retail routers implementing 802.11ac, unless one of the newer 802.11ac competitors can deliver MU-MIMO products by exploiting antenna diversity and DSP modulation talents that so far have been absent from this market. In any event, low-end vendors like Realtek will still be able to offer products with fewer features targeting entry-level retail routers and gateways.

As for the relative size of the Wave 1 and Wave 2 revolutions, all four chipset companies consider the distinction to be a short-term wrinkle—one likely to be forgotten by early 2016, when all major access points (and many client products) will have already upgraded to full Wave 2 support. The key to making such a transition cost effective, however, will be for all four vendors to support 160MHz channels in their MU-MIMO designs.


Price and Availability

None of the four major 802.11ac vendors has announced pricing. The Qualcomm Atheros suite of 998x and 999x chipsets is in production. Broadcom’s radio chips and processors (BCM4366 and BCM47094) are sampling. Marvell’s Avastar 88W8964 is sampling, and Armada 88F6282 is in production. Quantenna’s QSR1000 is in full production, but its 10Gbps product has yet to sample to customers.


Linley IoT Hardware Conference 2017
Focusing on hardware design for the IoT and wearables
July 25 - 26, 2017
Hyatt Regency Hotel, Santa Clara, CA
Register Now!
More Events »


Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »